DeepSeek is the Chinese AI lab whose R1 reasoning model stunned the industry by matching frontier performance at a fraction of the training cost, then giving the weights away. Its free assistant and radically cheap API made top-tier AI accessible to everyone.

DeepSeek offers a free chat assistant on web and mobile powered by its V-series general models and R-series reasoning models. The reasoning models show their chain of thought openly, thinking through math, logic, and code step by step before answering, which made DeepSeek-R1 a favorite for hard problems. The models are released with open weights, so the same intelligence runs on private infrastructure via Hugging Face or hosted providers.

For developers, DeepSeek's API undercuts Western frontier pricing by an order of magnitude, with OpenAI-compatible endpoints that make switching nearly effortless. That combination, frontier-class reasoning, open weights, and commodity pricing, permanently changed the economics of the model market.

Who is it for?

Developers and startups optimizing inference costs, researchers who need open weights, students who want free frontier reasoning, and anyone benchmarking against the big Western labs. Privacy-sensitive users should note data flows through Chinese infrastructure on the hosted service, or self-host instead.

How much does DeepSeek cost?

The assistant is completely free. API pricing is usage-based per million tokens at rates far below Western frontier models, with off-peak discounts; self-hosting via open weights costs only your compute.

Our verdict

DeepSeek is the value disruptor of frontier AI. Hosted-service privacy tradeoffs are real, but between the free app, open weights, and API pricing, it forces every serious AI user to at least consider it.